[ValueTracking] Use guards to prove non-nullness of a value

[reames]

test/Transforms/LICM/hoist-deref-load.ll

531

You're missing the CHECKs need to prove it doesn't trigger.

[sanjoy]

test/Transforms/LICM/hoist-deref-load.ll

531

Good catch, fixing.

[sanjoy]

• fix test

[reames]

LGTM

lib/Analysis/ValueTracking.cpp

for (auto *U : V->users()) {
     // Consider only compare instructions uniquely controlling a branch
     CmpInst::Predicate Pred;
     if (!match(const_cast<User *>(U),
                m_c_ICmp(Pred, m_Specific(V), m_Zero())) ||
         (Pred != ICmpInst::ICMP_EQ && Pred != ICmpInst::ICMP_NE))
       continue;
 
     for (auto *CmpU : U->users()) {
-      const BranchInst *BI = dyn_cast<BranchInst>(CmpU);
-      if (!BI)
-        continue;
-
+      if (const BranchInst *BI = dyn_cast<BranchInst>(CmpU)) {
         assert(BI->isConditional() && "uses a comparison!");
 
         BasicBlock *NonNullSuccessor =
             BI->getSuccessor(Pred == ICmpInst::ICMP_EQ ? 1 : 0);
         BasicBlockEdge Edge(BI->getParent(), NonNullSuccessor);
         if (Edge.isSingleEdge() && DT->dominates(Edge, CtxI->getParent()))
           return true;
+      } else if (Pred == ICmpInst::ICMP_NE &&
+                 match(CmpU, m_Intrinsic<Intrinsic::experimental_guard>()) &&
+                 DT->dominates(cast<Instruction>(CmpU), CtxI)) {
+        return true;
+      }
     }
   }
 
   return false;
 }
 
 bool llvm::isKnownNonNullAt(const Value *V, const Instruction *CtxI,
                    const DominatorTree *DT, const TargetLibraryInfo *TLI) {

test/Transforms/LICM/hoist-deref-load.ll

for.inc:                                          ; preds = %for.body, %if.then
   %lftr.wideiv = trunc i64 %indvars.iv.next to i32
   %exitcond = icmp eq i32 %lftr.wideiv, %n
   br i1 %exitcond, label %for.end, label %for.body
 
 for.end:                                          ; preds = %for.inc, %entry
   ret void
 }
 
+declare void @llvm.experimental.guard(i1, ...)
+
+define void @test12(i32* noalias %a, i32* %b, i32* dereferenceable_or_null(4) %c, i32 %n) #0 {
+; Prove non-null ness of %c via a guard, not a branch.
+
+; CHECK-LABEL: @test12(
+entry:
+  %not_null = icmp ne i32* %c, null
+  call void(i1, ...) @llvm.experimental.guard(i1 %not_null) [ "deopt"() ]
+  %cmp11 = icmp sgt i32 %n, 0
+  br i1 %cmp11, label %for.body, label %for.end
+
+; CHECK: for.body.preheader:
+; CHECK-NEXT:  [[VAL:%[^ ]]] = load i32, i32* %c, align 4
+; CHECK-NEXT:  br label %for.body
+
+
+for.body:                                         ; preds = %entry, %for.inc
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc ], [ 0, %entry ]
+  %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
+  %0 = load i32, i32* %arrayidx, align 4
+  %cmp1 = icmp sgt i32 %0, 0
+  br i1 %cmp1, label %if.then, label %for.inc
+
+if.then:                                          ; preds = %for.body
+  %1 = load i32, i32* %c, align 4
+  %arrayidx3 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv
+  %2 = load i32, i32* %arrayidx3, align 4
+  %mul = mul nsw i32 %2, %1
+  store i32 %mul, i32* %arrayidx, align 4
+  br label %for.inc
+
+for.inc:                                          ; preds = %for.body, %if.then
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.inc, %entry, %entry
+  ret void
+}
+
+define void @test13(i32* noalias %a, i32* %b, i32* dereferenceable_or_null(4) %c, i32 %n) #0 {
+; Like @test12, but has a post-dominating guard, which cannot be used
+; to prove %c is nonnull at the point of the load.
+
+; CHECK-LABEL: @test13(
+entry:
+  %not_null = icmp ne i32* %c, null
+  %cmp11 = icmp sgt i32 %n, 0
+  br i1 %cmp11, label %for.body, label %for.end
+
+; CHECK: for.body.preheader:
+; CHECK-NOT:  load i32, i32* %c
+; CHECK:  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.inc
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc ], [ 0, %entry ]
+  %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
+  %0 = load i32, i32* %arrayidx, align 4
+  %cmp1 = icmp sgt i32 %0, 0
+  br i1 %cmp1, label %if.then, label %for.inc
+
+if.then:                                          ; preds = %for.body
+  %1 = load i32, i32* %c, align 4
+  %arrayidx3 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv
+  %2 = load i32, i32* %arrayidx3, align 4
+  %mul = mul nsw i32 %2, %1
+  store i32 %mul, i32* %arrayidx, align 4
+  br label %for.inc
+
+for.inc:                                          ; preds = %for.body, %if.then
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.inc, %entry, %entry
+  call void(i1, ...) @llvm.experimental.guard(i1 %not_null) [ "deopt"() ]
+  ret void
+}
+
 attributes #0 = { nounwind uwtable }
 !0 = !{i64 4}